Smart City

Overview

From Pilot to Citywide AI Public Safety

A Tokyo metropolitan ward of roughly 210,000 residents has become one of the first municipalities in Japan to deploy AI-equipped security cameras at the ward-government level. The ward already operates a sizable existing camera network across its streets and stations; a tranche of those cameras around the highest-traffic transit hubs is being upgraded with on-camera AI inference, alongside additional battery-backed disaster cameras deployed along arterial roads.

Atomrock provides the underlying stack: edge AI devices at each camera location, a central management cloud, and a unified control interface at the Ward Office. Real-time monitoring is integrated with police and fire response systems, while privacy controls limit live access to authorized staff and enforce a defined retention period.

Challenge

Citywide Safety Without Citywide Operators

Like much of Tokyo, the ward faces a converging set of public-safety pressures. An aging population means more dementia patients and vulnerable elderly residents going missing on the streets. High-traffic transit hubs see growing pedestrian flows. Disaster preparedness, earthquake response, typhoon evacuation, road-condition monitoring, has become a permanent municipal mandate, not a periodic drill.

The existing camera infrastructure recorded incidents but did not act on them. The ward needed a real-time AI layer running on top of its existing network, search-and-locate capability for missing persons, evacuation-grade reliability for disaster scenarios, and a privacy posture credible enough to satisfy residents — without expanding the operator headcount of the public safety team.

Solution

Distributed Edge AI on Existing Camera Networks

Atomrock deployed a four-tier architecture: cameras feed into edge AI devices, edge devices stream to the Atomrock central management cloud, and authorized staff at the Ward Office work from a unified central control interface. Facial recognition and person-tracking inference runs on the edge, so identification happens locally with minimal latency, and only the relevant events and metadata are pushed upstream.

The system integrates directly with police and fire department dispatch workflows. When a missing-person search starts at the Ward Office or an incident is flagged on patrol, the same console drives the response. Disaster cameras with battery backups stay online when grid power doesn't, providing live road-condition footage that the ward uses to direct evacuations along arterial streets.

Privacy is enforced by design: footage is automatically deleted after a defined retention window, and live viewing is restricted to a small set of authorized ward staff. Centralized management, distributed execution — one operator console covers the entire ward without expanding the surveillance surface in residents' daily lives.

Result

Faster Searches, Coordinated Disasters, Stronger Public Trust

In production, the system delivers across the three pillars the ward originally framed as the public-safety mandate: crime prevention through faster identification and tracking of suspects across the camera network; disaster response through live road and evacuation visibility from cameras that stay online when the grid doesn't; and community protection through a search capability that helps locate missing children and dementia patients in minutes rather than hours of street-level patrol.

Investigation efficiency at the police interface is up. Response coordination between ward, police, and fire works from a single shared console. Crucially, the privacy posture — defined retention, restricted access, edge-local inference — has held up under public scrutiny, which is the precondition for any AI public-safety platform to expand. This ward is the proof point; further Tokyo districts are evaluating expansion onto the same architecture.

To learn how Atomrock's AI public safety platform can support municipal, transit, and civic-infrastructure deployments, contact us.